机械油泵轴锻压工艺的正则化递归ESN网络优化  

Optimization on Forging Process of Mechanical Oil Pump Shaft Based on Norm Recursive ESN Network

作　　者：徐安林[1] 钱善华 XU An-lin;QIAN Shan-hua(l.Wuxi Institute of Technology,Jiangsu Wuxi 214121,China;School of Mechanical Engineering,Jiangnan University,Jiangsu Wuxi 214122,China)

机构地区：[1]无锡职业技术学院,江苏无锡214121 [2]江南大学机械工程学院,江苏无锡214122

出　　处：《机械设计与制造》2022年第10期85-88,93,共5页Machinery Design & Manufacture

基　　金：江苏省食品先进制造装备技术重点实验室自主研究课题项目(FMZ201907)。

摘　　要：为了提高机械油泵轴的耐磨性、减少油泵轴工作过程中的磨损量,提出了基于L1正则化递归最小二乘ESN网络的锻压工艺优化方法。分析了油泵轴的生产工艺流程和磨损实验方法。以回声状态网络为基础,为了解决训练过程中动态储备池中大量无关特征问题,将L1范数正则化约束加入到算法中,使ESN网络对特征具有自主选择能力,实现动态储备池中特征稀疏性,解决ESN网络计算量大和过拟合问题;使用递归最小二乘法优化输出权值矩阵,可以实时接受新样本对模型精度的训练和提高,综合以上两点创新,提出了L1RLS-ESN网络模型。将L1RLS-ESN网络应用于油泵轴磨损量的拟合与预测,得到了磨损量最小的对应锻压工艺。将模型优化的锻压工艺和实际锻压工艺进行对比,模型优化工艺锻压的泵轴磨损量比实际锻压工艺减少了约2倍。To improve wear-ability of mechanical oil pump shaft,and lessen wear extent during the operation of oil pump shaft,forging technology optimization method based on L1 regularization and recursive least squares ESN network is proposed.Producing technology process of oil pump shaft is introduced and wearing experiment of it is designed.on the basis of echo state network,to reduce irrelevant feature in reserve pooling producing during training process,L1-norm regularization constraint is introduced to the algorithm,so that ESN network possess choosing ability independently on feature.Feature sparsity could come true,and heavy calculation burden and overfitting problem of ESN network is solved.Outputting weight matrix is optimized by recursive least squares,so that new sample could train the model and improve its accuracy real-time.Synthesizing the above two innovation,L1RLS-ESN network model is put forwarded.L1RLS-ESN network model is used to fit and predictive wearing extent of oil pump shaft,optimal forging technology corresponding to minimum wearing extent is gotten.Comparing the wearing extent of oil pump shaft produced by optimized forging technology and real forging technology,wearing extent of oil pump shaft produced by optimized forging technology decreases by 2 times compared with pump shaft produced by real forging technology.

关 键 词：机械油泵轴 锻压工艺优化 L1RLS-ESN网络 L1范数正则化 递归最小二乘法 

分 类 号：TH16[机械工程—机械制造及自动化] TG316[金属学及工艺—金属压力加工]